Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Factor structure and invariance of the CES-D 1
Running Head: FACTOR STRUCTURE AND INVARIANCE OF THE CES-D
Factor Structure of the CES-D and Measurement Invariance Across Gender in Mainland
Chinese Adolescents
Mengcheng Wang
Central South University, Changsha, China
Cherie Armour
University of Southern Denmark, Funnen, Denmark
Yan Wu
Guangdong University of Foreign Studies, Guangzhou, China
Fen Ren
Institute of Psychology, Chinese Academy of Sciences, Beijing, China
Xiongzhao Zhu, and Shuqiao Yao
Central South University, Changsha, China
Author Note
Mengcheng Wang, Xiongzhao Zhu, and Shuqiao Yao, Medical Psychological Institute,
Second Xiangya Hospital, Central South University, Changsha, China; Cherie Armour,
National Centre for Psychotraumatology, University of Southern Denmark, Odense Campus,
Funnen, Denmark. Yan Wu, Department of Applied Psychology, Guangdong University of
Foreign Studies, Guangzhou, China. Fen Ren, Key Laboratory of Mental Health, Institute of
Psychology, Chinese Academy of Sciences, Beijing, China.
Correspondence concerning this article should be addressed to Shuqiao Yao, Medical
Psychological Institute, Second Xiangya Hospital, Central South University, #139 Ren-Min
Zhong Road, Changsha, 410011, China. Phone: 86-731-85292130, FAX: 86-731-85361328,
E-mail: [email protected]
Factor structure and invariance of the CES-D 2
Abstract
Objectives: The primary aim was to examine the depressive symptom structure of mainland
China adolescents using the Center for Epidemiologic Studies Depression Scale (CES–D).
Design: Exploratory factor analysis and confirmatory factor analysis were simultaneously
conducted to determine the structure of the CES-D in a large scale, representative adolescent
samples recruited from mainland China. Multiple-group confirmatory factor analysis (N =
5059, 48% boys, M = 16.55±1.06) was utilized to test the factorial invariance of the
depressive symptom structure which was generated by EFA and confirmed by CFA across
gender. Results: The CES-D can be interpreted in terms of three symptom dimensions.
Additionally, factorial invariance of the new proposed model across gender was supported at
all assuming different degrees of invariance. Conclusion: Mainland Chinese adolescents
have specific depressive symptom structure which is consistent across gender.
Key Words: Depressive symptom; CES–D; factor structure; factorial invariance; Chinese
adolescent
Factor structure and invariance of the CES-D 3
Factor Structure of the CES-D and Measurement Invariance Across Gender in Mainland
Chinese Adolescents
The Center for Epidemiologic Studies Depression Scale (CES-D: Radloff, 1977) is a
commonly employed self-report measure of depressive symptomatology. The CES-D was
originally designed and developed for the identification of depression among the general
adult population. The scale items cover the major components of depressive symptomatology
and were selected from previously validated depression scales. The CES-D has since been
utilized across various age groups, including elderly people (e.g., Zhang, Fokkema, Cuijpers,
Smits, & Beekman, 2011), adult community samples (e. g., Roberts, & Vernon, 1983), college
students (e. g., Yen, Robins, & Lin, 2000), adolescents (e.g., Cheng, Yen, Ko, & Yen, 2012;
Lee et al., 2008; Radloff, 1991; Roberts, Andrews, Lewinsohn, & Hops, 1990), and children
(e.g., Li, Chung, & Ho, 2010).
Radloff (1977) originally proposed that the 20 CES-D items were categorized into four
symptom groups; Depressed Affect (DA, 7 items), Somatic Complaints (SC, 7 items),
Interpersonal Problems (IP, 2 items), and Positive Affect (PA, 4 items). Subsequent support
for the four-factor model has been plentiful across various populations from the United States
(e.g., Golding, & Aneshensel 1989; Hertzog, Van Alstine, Usala, Hultsch, & Dixon, 1990;
Roberts et al, 1990; Rozario, & Menon, 2010; Roberts & Vernon, 1983). However, a number
of other studies employing exploratory factor analysis (EFA) and confirmatory factor analysis
(CFA) have yielded different CES-D factor structures, particularly among populations that
are demographically distinct from those where the CES-D was developed. To date, studies
have reported two (Cheung & Bagley, 1998; Edman, Danko, Andrade, McArdle, Foster, &
Glipa, 1999; Kazarian, & Taher, 2010; Rivera-Medina, Caraballo, Rodrıguez-Cordero, Bernal,
& Davila-Marrero, 2010), three (Guarnaccia, Angel, & Worobey, 1989; Kuo, 1984; Liang,
Factor structure and invariance of the CES-D 4
Van Tran, Krause, & Markides, 1989; Yen, Robins, & Lin, 2000; Ying, 1988), and, as many
as, five (Thorson & Powell, 1993; Ying, Lee, Tsai, Yeh, & Huang, 2000) factors. Furthermore,
some studies have reported four-factor solutions that significantly differed from Radloff’s
original classification (Crockett, Randall, Shen, Russell, & Driscoll, 2005; Foley, Reed,
Mutran, & DeVellis, 2002).
Shafer (2006) conducted a relatively recent meta-analysis which investigated the CES-D
factor structure with over 22,000 participants, from 21 CES-D studies. Using EFA, the
meta-analysis concluded that the original Radloff four-factor structure was the best solution.
However, this study also described reasonable two and three-factor solutions, raising further
questions about the universality of the four-factor solution. More recently, Kim and
colleagues (2011) conducted a meta-analysis including 13 exploratory factor analytic studies
(N = 19, 206) and 16 confirmatory factor analytic studies (N = 65,554) that investigated the
structure of CES-D across various racial/ethnic groups. They concluded that CFA studies
supported the original four-factor structure across all racial/ethnic groups, with the exception
of Asian groups. However, results from the EFA meta-analysis did not consistently replicate
the original four-factor solution. Therefore, these authors (Kim et al., 2011) recommended
that researchers should first use EFA and secondly apply CFA when examining the factor
structure of the CES-D in racially/ethnically diverse groups. Thus, the current study shall
follow their recommendation. Moreover, it is worth noting that the meta-analysis only
encompassed seven studies (EFA = 4, CFA = 3) which utilized child and adolescent samples.
Consequently, more attention should be paid to the structure of the CES-D within child and
adolescent groups, particularly in non-western cultures.
To date, the results of previous CES-D factor analytic studies within Chinese samples
(both adult and adolescent) have been mixed (cf. Cheng et al., 2012; Lee et al., 2008; Li,
Chung, & Ho, 2010; Lin, Wei, Yi, Xiao, & Yao, 2008; Yen, Robins, & Lin, 2000; Zhang,
Factor structure and invariance of the CES-D 5
Fokkema, et al., 2011; Zhang, Wu, Fang, Li, Han, & Chen, 2010). Indeed, a study conducted
with mainland Chinese high school students (Lin et al., 2008) reported that the initial
four-factor model proposed by Radloff (1977) fitted the data appropriately, however they
failed to compare the four-factor model with other competing models. Similar results were
observed in a mainland general population study (Zhang, Wu, et al., 2010), a study utilizing
an elderly sample (Zhang, Fokkema, et al., 2011), Taiwan adolescents (Cheng et al., 2012),
and a sample of children from Hong Kong (Li, Chung, & Ho, 2010). However, an alternative
study based on Chinese university students, which employed principal components analysis
with varimax rotation (Yen, Robins & Lin, 2000) identified three factors. In this solution,
items from the original DA, IP, and SC factors merged into two new factors labeled as a
somatic factor and an affective factor. This model confirmed the results of previous studies
(Liang et al., 1989; Manson et al., 1990) and was recently confirmed in a confirmatory factor
analytic study utilizing data from Hong Kong adolescents (Lee et al., 2008).
Moreover, several studies have found that two to five factor solutions fit their data best.
In a sample of 138 Hong Kong married couples, Cheung and Bagley (1998) reported a
two-factor structure that separated the IP items from the remaining items, represented the best
fit to the data. Inconsistent with Cheung and Bagley’s two-factor solution, Lam et al. (2004)
reported an alternative two-factor solution, with the four positive items loading on a factor
separate from the remaining items, in a Hong Kong adolescent sample. However, Ying et al.
(2000) reported a five-factor solution in a Chinese American college student sample. Thus, a
consensus regarding the underlying dimensionality of the CES-D with the population of
China has yet to be reached.
Notably, there may be several limitations to the analytic approach used in previous
Chinese sample studies. First, the data collected from the CES-D has often been treated as
continuous despite items being based on only four response categories (Cheng et al., 2012;
Factor structure and invariance of the CES-D 6
Li, Chung, & Ho, 2010; Zhang, Fokkema, et al., 2011; Zhang, Wu, et al, 2010). Treating
Likert rating scale data, which has four, or less than four, response categories as continuous
outcomes violates the assumption of multivariate normality, and thus may distort the resultant
factor structure and associated parameters (DiStefano, 2002; Lubke & Muthén, 2004).
Second, most of the above-mentioned studies, which are based on Chinese samples,
employed CFA to compare several alternative models of the CES-D. This may reflect a
missed opportunity to uncover a Chinese specific structure through the use of EFA (Kim et al.,
2011). Third, although some researchers have tested the measurement invariance of the
CES-D between Chinese populations with varying cultural backgrounds (Zhang, Fokkema, et
al., 2011), no study has examined the measurement invariance of the CES-D across gender in
the mainland China population. Nor has such been implemented using an adolescent sample.
In order to overcome those shortfalls, the present study utilizes a representative mainland
Chinese adolescent sample, employing both EFA and CFA to explore the structure of the
CES-D.
Indeed, exploring the factor structure and measurement invariance of the CES-D has
significant meaning for both clinicians and researchers. From a cross-cultural standpoint,
depression is sometimes manifested and expressed differently across various cultural groups
and subgroups (e.g., Iwata et al., 2002), so exploring the factor structure of a depression
measure (e.g., CES-D) enables us to examine any differences across cultures and groups.
Moreover, different factor structures imply variant psychological mechanisms, thus
suggesting that diverse therapeutic schedules should be employed for differing people.
Specifically, effective therapeutic schedules verified in one culture may be invalid for people
from another culture.
The main aim of this study was to examine the factor structure of the CES-D in a large
representative mainland Chinese adolescent sample. Inconsistent with previous studies using
Factor structure and invariance of the CES-D 7
Chinese samples (e.g., Cheng et al., 2012; Zhang, Fokkema, et al., 2011) and according to the
recommendation of Kim et al. (2011), we first employed EFA to explore the structure of the
CES-D in sample one, and then in sample two we used CFA to compare the EFA derived
model with other competing models which were shown to be superior in previous studies.
Moreover, although no significant gender differences were found in Chinese adolescents
based on manifest mean-levels (Greenberger, Chen, Tally, Qi, 2000; Tepper, Liu, Guo, Zhai,
Liu, & Li, 2008; Yen, Robins, & Lin, 2000), if measurement invariance does not hold across
gender, differences in observed depressive symptoms scores may not be directly comparable
(Meade, Lautenschlager, & Hecht, 2005). Thus, the second aim was to examine measurement
invariance of the best fitting model, identified in above analysis, across gender, and to
compare sex differences of depressive symptoms at the latent mean level.
Method
Participants and Procedure
A total of 5059 participants (48% boys) were recruited whose ages ranged from 14 to 20
(M = 16.55, SD = 1.06). Among study participants, 47.7% were in the United States’
equivalent of Grade 9, with 30.2% in Grade 10, and 22.1% in Grade 11. The sample
encompassed 93.4% of individuals who reported their ethnicity as Han. A further 6.6%
classified themselves as belonging to an ethnic minority or declared that their ethnicity was
unknown. With regards to family composition, participants reported the following: 88.5%
declared that their family was a nuclear family, 7.1% reported divorced families, 3.2%
reported remarried families, and 2.2% reported single-parent families.
Participants completed the survey in school during a specified class period lasting
approximately 45 minutes. Informed consent was given by parents (or legal guardians) prior
to the administration of the self-reported questionnaires. Furthermore, children provided
Factor structure and invariance of the CES-D 8
assent for their own participation. Ethical approval was given by the Human Subjects Review
Committee at Central South University.
Measure
The Chinese version of the CES-D. The CES-D was used to measure the level of
depressive symptomatology. The CES-D consists of 16 negative affect and 4 positive affect
items, such as “I felt depressed”, “I felt lonely”, and “I was happy”. Participants were asked
about the number of days on which they experienced depressive symptoms during the
previous week. Respondents reported the frequency of occurrence, of each item, during the
previous week on a 4-point scale: 0 (rarely; less than 1 day), 1 (some of the time; 1–2 days), 2
(a moderate amount of the time; 3–4 days), or 3 (most or all of the time; 5–7 days). Higher
scores on the CES-D indicate more depressive symptoms (Radloff, 1977). The four positive
affects items were reversed before conducting our analysis. The Chinese version of this scale
has been validated (Cheng et al., 2012; Lee et al., 2008; Zhang, Wu, et al., 2010) and
extensively utilized in Chinese studies (e.g., Yen, Robins, & Lin, 2000; Zhang, Fokkema, et
al., 2011; Zhang, Wu, et al., 2010).
Data analysis strategy
Missing Data
The original sample included 5321 adolescents, however given that 262 failed to respond
to all of the CES-D items they were excluded from the analysis. This left an effective sample
size of 5059.
Analytic Steps
Our analyses contained three steps. First, EFA was conducted on a randomly split half of
the whole sample (n =2526). EFA was used to identify the best fitting factor model of the
CES-D, in the present sample. Subsequently, we used a random split half (n = 2533) of the
Factor structure and invariance of the CES-D 9
sample to run CFA. The CFA tested the fit of a number of competing models; including the
model which was generated from our EFA. Finally, utilizing the full sample, we assessed
measurement invariances of the best fitting model, from the CFA, across gender.
Stage 1: Exploratory Factor Analysis
Descriptive statistics and EFA were performed by the SPSS program (IBM, SPSS version
17, 2009). In line with previous studies (e.g., Radloff, 1977; Yen, Robins, & Lin, 2000),
principal components analysis with varimax rotation was performed to determine the number
of factors. However, given that the oblique rotation approach is also appropriate in this case,
we computed the oblique rotation solution. Notably, the factor loading matrix was similar to
that of the varimax rotation solution. Further details and matrices are available from the first
author.
Stage 2: Confirmatory Factor Analysis
A series of confirmatory factor analyses were specified and estimated using Mplus 5.1
software (Muthén & Muthén, 1998–2007). Given that items have only four response
categories, Maximum Likelihood (ML) estimation was deemed inappropriate in light of
simulations studies showing that a minimum of five response categories are a prerequisite to
the assumptions of continuity underlying ML estimation (DiStefano, 2002; Lubke & Muthén,
2004). Thus, the robust weighted least-squares with mean and variance adjustment (WLSMV)
estimator was used in present study (Flora & Curran, 2004).
CFA Model Specification
Five alternative models of the CES-D, which were shown to be best fitting across
previous studies were chosen for comparison (cf. Table 1). Model 1 tested Radloff’s original
four-factor model in which all 20 items loaded on four specific factors: depressed, somatic,
interpersonal, and positive. The four-factor model has been shown to be the best fitting model
across various Chinese factor analytic studies (Cheng et al., 2012; Lee et al., 2008; Li, Chung,
Factor structure and invariance of the CES-D 10
& Ho, 2010; Zhang, Fokkema, et al., 2011; Zhang, Wu, et al., 2010). Model 2 tested a two
factor model identified in Cheung and Bagley’ study (1998) in which the interpersonal
problems items were separated from the remaining items. Model 3 tested another two factor
model that combined all negative items into an independent factor whilst the remaining
positive items formed a second factor. Recently, this particular model has been verified as
superior in several studies (Edwards, Cheavens, Heiy, & Cukrowicz, 2010; Kazarian, & Taher,
2010; Leykin, Torres, Aguilera, & Muñoz, 2011; Rivera-Medina et al., 2010; Schroevers et al.,
2000). Notably, it has been shown to be superior in Hong Kong (Lam et al., 2004) and
Filipino-American (Edman et al., 1999) adolescents. Model 4 tested the 3 factor model that
was identified in EFA (see EFA results section). Model 5 tested an alternative three factor
model encompassing a factor of depressed affect and somatization, a factor of positive affect,
and a factor of interpersonal problems (Guarnaccia et al., 1989; Kuo, 1984; Ying, 1988).
Model evaluation in CFA
Following generally accepted practice, we evaluated the fit of each model by examining
multiple fit indices (Kline, 2010). We used the Chi-square, root-mean-square error of
approximation (RMSEA), Tucker-Lewis index (TLI) and comparative fit index (CFI).
Conventional guidelines suggest that RMSEA values ≤ .08 indicate acceptable model fit and
≤ .05 indicate good model fit, and CFI, TLI ≥ .90 indicate adequate model fit (Kline, 2010).
Regular chi-square difference tests were not conducted here for the comparison of
non-nested competing models. Rather, we employed the Bayesian information criterion (BIC)
as it was most suitable in this case. A 10-point BIC difference represents a 150:1 likelihood (p
< .05) that the model with the lower BIC value fits best; a difference in the 6-10 point range
indicates “strong” support, and > 10 indicates “very strong” support (Raftery, 1995).
However, given that the BIC value is not given when utilizing the WLMSV estimator in
Mplus we computed the BIC value by estimating the models using ML estimator.
Factor structure and invariance of the CES-D 11
Stage 3: Measurement Invariance
After identification of the best fitting model of the CES-D, we conducted measurement
invariance tests across gender. Measurement invariance tests were performed using the
sequential strategy described by Meredith and Teresi (2006). First, Model A tested configural
invariance by allowing all parameters to vary. Subsequent models tested sequentially more
conservative restrictions of Model A by constraining particular parameter estimates to be
equal across groups. Subsequent models were tested against the prior step’s model (except
when noted otherwise). Model B constrained factor loadings across groups (testing weak
invariance). Model C additionally constrained observed variable thresholds (testing strong
factorial invariance). Model D additionally constrained residual error variances (testing strict
factorial invariance). Model E, tested structural invariance by additionally constraining factor
variances and covariances (but not error variances), tested against Model C; and finally,
Model F, additionally constrained factor means (but not residual variances), while being
tested against Model E. Thus, we not only tested differences in symptom structure between
the boys and girls, but also differences in the level of symptom severity.
Model comparison in Measurement Invariance
Given that previous research has reported that comparing nested models solely on the
bases of fit indices may lead to inaccurate conclusions (Fan & Sivo, 2009), nested models in
the current study were compared using the difftest function available in Mplus (Muthén &
Muthén, 1998–2007). However, given that tests of the change in CFI are reported as being
superior to chi-square difference tests of nested models, because they are not affected by the
sample size (Cheung & Rensvold, 2002; Meade, Johnson, & Braddy, 2008), the current study
compared nested models in consideration of both the chi-square difference test and CFI
values. According to Cheung and Rensvold (1999, 2002), a CFI difference of < .01 indicates
that the invariance hypothesis should not be rejected as mean differences exist when CFI
Factor structure and invariance of the CES-D 12
differences are .01 to .02; and definite differences exist when CFI differences are > .02
(Cheung & Rensvold, 1999, 2002).
Results
Stage 1: Exploratory Factor Analysis
According to the suggestion offered by Kim et al.(2011), EFA was conducted firstly on a
random split half of whole data-set to choose the best factor model representing the structure
of CES-D in present population. This analysis yielded 3 factors/components with eigenvalues
greater than 1.0, moreover, the scree plot test also suggested the retention of three factors.
The three factors cumulatively accounted for 48.58% of the total variance. As can be seen in
table 1, the first factor explained 19.89% of the total variance and consists of nine items that
belong to the original DA (4 items: 3, 6, 9, 10) and SC (6 items: 1, 2, 5, 7, 11, 13) factors, as
well as two cross-loading items 9 and 18. The second factor explained 16.95% of the total
variance and consists of 7 items that belong to the original DA (3 items: 14, 17, 18), IP (2
items: 15, 19), and SC (2 items: 13, 20) factors. The last factor explained 11.74% of the total
variance and consists of 4 positive affect items. These results are consistent with findings
from other studies (Liang et al., 1989; Yen, Robins, & Lin, 2000; Manson et al., 1990).
The 3 factor structure found in the EFA in the current study differs from Radloff’s (1977)
factors. In particular, the IP factor dropped out in this study. Additionally, four of seven DA
items loaded onto the SC factor and two of seven DA items loaded onto a new DA factor. It
was worth noting that the two items, item 9 and item 18, cross-loaded onto more than one
factor. Item nine primarily loaded (loading size = .473) onto SC factor and peripherally
loaded onto two other factors (factor loadings were .378 and .376, respectively). Item 18
primarily loaded (factor loading = .559) onto the SC factor and peripherally loaded onto the
DA factor (factor loading = .503). Notably, a previous study found similar results (i.e. Yen,
Factor structure and invariance of the CES-D 13
Robins, & Lin, 2000).
Stage 2: Confirmatory Factor Analysis
Table 3 summarizes the fit indices of these competing models using the WLSMV
estimation method for the polychoric correlation matrix of the CES-D items in the second
split half of the full sample. As can be seen in table 3, all of the examined models fit the data
well (CFIs > .90, TLIs > .90, RMSEAs < .08 ) with the exception of Model 2 (WLSMVχ2 =
5350.953, CFI = .887 < .90, TLI = .873 < .90, RMSEA = .11 > .08, BIC = 100941.619) in
which IP items were separated from the remaining items. Overall, Model 5 provided the best
fit to the data among the five alternative models (WLSMVχ2=2179.399, df = 167, CFI = .956,
TLI = .950, RMSEA = .069, BIC= 99862.198), in which the DA factor and the SC factor
were collapsed into single factor, in addition to PA and IP factors. The second best fitting
model was Radloff’s original four-factor model (WLSMVχ2=2183.48, df = 164, CFI = .956,
TLI = .949, RMSEA = .070, BIC= 99875.173). The difference in the resultant BIC value
between Model 5 and Radloff’s model was 12.975 (> 10), thus suggesting that Model 5 fit the
data significantly better than Radloff’s model.
The model identified in the above EFA, Model 4, also showed acceptable fit to the data.
However, given that Model 4 included two cross-loading items (i.e., item 9 and item 18) we
hypothesized this may result in worse fit therefore we checked the modification indices (MI)
of M4. Indeed, the MI’s indicted that if item 9 loaded onto the PA factor and item 18 loaded
onto the SC factor, the chi square value would reduce 267.102 and 126.477, respectively.
Thus, based on the MI’s we allowed item 9 to load onto the PA factor and item 18 to load
onto the SC factor. We then estimated this model using CFA. The fit indices of the modified
model (M4-modified) indicated excellent fit, (CFI = .963, TLI = .957, RMSEA = .064, BIC =
99797.885), albeit with a significant chi-square value, WLSMVχ2 (165, N = 2530) =
1866.193, p < .001. All factor loadings were significant at p < .001.
Factor structure and invariance of the CES-D 14
We also employed an alternative analytic procedure to allow for a simple structure
and lucid explainable results; thus, we created a new version of model 4: M6. This model was
essentially identical to the above however we excluded items nine and 18. However, to
ensure we were following correct procedures we firstly calculated the pearson
product-moment correlation between original factor scores and new factor scores without the
two items. The coefficients were .990 for SC, and .986 for DA respectively, suggesting that
the two removed items did not attenuate the correlations between factors. We once again
estimated this new model, model M6 using CFA. The fit indices of M6 exhibited an excellent
fit (WLSMVχ2 =1354.121, df = 132, CFI = .966, TLI = .961, RMSEA = .060, BIC =
91410.876). Solely from the point of model fitting, M6 was the best fitting model compared
to other alternative models. However, the difference in fit between M6 and the modified M4
was minimal, but the differences between Model 4 and Model 6 in terms of the resultant BIC
values, was 8387.007 ( > 10), thus suggesting that Model 6 fit the data significantly better
than Model 4.
Stage 3: Factorial Invariance Across Gender
The results (see Table 4) from the measurement invariance revealed that all of the five
steps of invariance testing resulted in significant χ2 (all ps < .01), excellent (CFIs > .95,
TLIs > .95) and equivalent fit indices (∆CFIs < .01, ∆TLIs < .01), with significant ∆χ2 (all ps
< .01 expect for Model E, p < .05). All goodness-of-fit indices, except for χ2 and Δχ2, which
are sensitive to large sample size, indicated that all models assuming different degrees of
invariance were acceptable. In summary, these results suggest that the CES-D items have the
same meanings across gender, manifest mean-levels (i.e., based on summed/averaged scores),
and latent mean comparisons, suggesting that comparisons across sex are meaningful.
Discussion
Factor structure and invariance of the CES-D 15
The primary purpose of this study was to test the factor structure of the CES-D using
EFA and CFA among mainland China adolescents. EFA results revealed that the twenty items
of the CES-D can be interpreted in terms of three symptom dimensions, including SC, DA,
and PA. This result is partly consistent with EFA findings from other studies (Yen, Robins, &
Lin, 2000; Manson et al., 1990), which suggest original three dimensions (SC, IP and DA)
collapsing into two factors: SC and DA. However, this finding is inconsistent with previous
researches who have employed CFA to compare alternative structures of depressive
symptoms among Chinese populations (Cheng, Yen, et al., 2011; Zhang, Fokkema, et al.,
2011).
To validate the structure of the CES-D confirmed in the first split half of the sample
using EFA, we specified and estimated several alternative models which have been
previously reported as best fitting across a number of studies. Models were tested using CFA
in the second split half of the sample. The EFA model which we found was similar to another
three factor model (Lee et al., 2008; Liang et al., 1989; Manson et al., 1990; Yen, Robins &
Lin, 2000) in which DA and SC merged into a single factor. However, this model included
two cross factor loadings and thus when we tested it using CFA it fit adequately but not
excellently. Therefore, we removed the items which had cross-factor loading and re-specified
the model. This newer modified model was found to fit the data best, compared to alternative
models, and compared to the model found in EFA which had cross-factor loadings. However,
to allow for simple structure and lucid explainable result, we planned to delete those
cross-loading items (i.e., item 9, 18). Thus a new model labeled M6 was tested, which
resulted in best fit. To guarantee this decision, we calculated the pearson product-moment
correlation between factor scores with the two items and scores without them and another
EFA. Consequently, a three factors model best represented the structure of the 18 items of
CES-D among Chinese adolescents.
Factor structure and invariance of the CES-D 16
Furthermore, the findings of the EFA and CFA together suggests that there is a specific
structure of depressive symptoms in mainland China adolescents which differs from the
structure found by recent alternative studies of Chinese adolescents (Cheng et al., 2012; Lin
et al., 2008; Zhang, Wu, et al., 2010). The structure also differs from that found in other
adolescent populations from varying cultural backgrounds (Edman et al., 1999; Radloff, 1991;
Roberts, & Sobhan, 1992). As mentioned above, prior studies (Cheng et al., 2012; Li, Chung,
& Ho, 2010; Lin et al., 2008; Zhang, Fokkema, et al., 2011; Zhang, Wu, et al., 2010 ) which
have been conducted using Chinese populations, have failed to explore the structure of
depressive symptoms prior to conducting the CFA (Kim et al., 2011). Additionally, CFA’s
have employed inappropriate estimators when conducting the CFA (e.g., ML). Notably, a
recent meta-analysis (Kim et al., 2011) proposed that when testing the structure of depressive
symptoms in new cultures, researchers should first implement EFA and then move to CFA.
Indeed, they suggested that conducting CFA as a first point of call may be misleading.
Moreover, although the original four-factor model of the CES-D provided reasonable fit to
the data using CFA in the current study; it was not found in the EFA. Consequently, our
results together with Kim et al.’s (2011) suggestion support the implementation of EFA prior
to CFA in the current study.
Our results also showed that the three-factor model which merged DA and SC factors
into one single factor, in addition to PA, and IP factors, fit the data best (that is prior to the
testing of modified models). The fact that DA and SC factors intertwine, implies that
mainland China adolescents are prone to mingle somatization with the expression of
depressive mood. In other words, Chinese adolescents tend to express somatic symptoms
more so than psychological symptoms of depression as more DA items were treated as SC
items. This result is in agreement with traditional cultural views that Chinese people are
encouraged not to express their mood condition, i.e., psychological symptoms; thus
Factor structure and invariance of the CES-D 17
somatization is an alternative way to express emotional disturbance (Kleinman, 1982;
Cheung, 1995). Furthermore, given that mental illnesses are stigmatized in Chinese societies
(Chan & Parker, 2004; Chung & Wong, 2004; Parker, Gladstone, & Chee, 2001; Ryder, Bean,
& Dion, 2000), this may further discourage the expression of psychological symptoms.
Consequently, somatic symptoms are emphasized in Chinese people (Cheung, 1995; Parker,
Gladstone, & Chee, 2001; Ryder et al., 2008). Interestingly, despite rapid modernization in
mainland China across the last decade, the way in which mental illness is expressed,
particularly depression, has not changed (Kleinman, 1982).
The divergence among studies with respect to the structure of the CES-D in Chinese
populations may be partially attributed to culture influence. Ying et al.’ (2000) study revealed
that SC and DA were separate factors in a sample of Chinese-American college student (n =
353), of which 34.56% were American-born and the rest were immigrants. However, in
studies of other Chinese populations, i.e., those who were born and raised in China (Yen et al.,
2000), and those who born in China and subsequently immigrated to the United States (Ying,
1988); results have concluded that the SC and DA items tend to mix. To the best of our
knowledge, until now, no one study has directly addressed this question. Thus, future
cross-cultural studies, considering explicit cultural variables (e.g., individualism vs.
collectivism; cultural identity), are needed in order to determine the cultural boundaries of the
model.
The second aim of the current study was to examine measurement invariance across
gender and to compare the sex difference of depressive symptoms on latent mean level.
Previous studies have found that sex has influence on the factor structure of the CES-D
(Clark, Aneshensel, Frerichs, & Morgan, 1981; Guarnaccia, Angel, & Worobey, 1989). In the
current Chinese adolescent sample, however, we failed to replicate this result. All models
assuming different degrees of invariance were acceptable, suggesting that the CES-D’s
Factor structure and invariance of the CES-D 18
factors have the same meaning across gender, as well as manifest mean-levels (i.e., based on
summed/averaged scores). Thus, suggesting that comparisons across sex are meaningful. Our
findings were consistent with Cheng et al.s’ (2012) investigation, in which no measurement
non-invariance were found, in a sample of Taiwan adolescents. As mentioned above, no
significant gender differences were reported in Chinese adolescents with regard to manifest
mean-levels (Greenberger et al., 2000; Tepper et al., 2008; Yen, Robins, & Lin, 2000), and
thus our findings supply additional empirical support for their conclusion. It is worth noting
that previous studies have compared the depressive symptoms of Chinese people with people
from various cultural backgrounds (e.g., Greenberger et al., 2000; Parker, Gladstone, & Chee,
2001; Ryder et al., 2008; Yen, Robins, & Lin, 2000). However, as related to gender, if the
measurement invariance does not hold across groups, differences in observed scores may not
be directly comparable. Thus, it is crucial that future studies take measurement invariance
into consider when conducting cross-cultural research.
Our results also have crucial meaning for clinicians. Firstly, mainland China adolescents
are prone to mingle somatization with the expression of depressive mood, which remind the
clinicians to be cautious with adolescent outpatients and to be wary that they do not obscure
and/or miss potential depressed adolescents. Secondly, specific therapeutic schedules should
be developed and employed for this group. Finally, it is significant that future depression
measurement and diagnostic criteria should take this character into consider.
The current study is not without limitation. For example we employed a restricted sample
of urban adolescents, thus the results may not fully generalize to all Chinese adolescents, in
particular those from rural parts of China. Furthermore, our sample was not a clinical one and
therefore may not generalize well to individuals who receive clinical diagnoses of depression.
More positively however, our study employed a large scale, representative urban sample. In
addition, we used more appropriate parameter estimate approach (Flora & Curran, 2004).
Factor structure and invariance of the CES-D 19
These aspects will have enhanced the research relevance of our results.
Conclusion
The results of the present investigation indicated that the original 4-factor model of the
CES–D was unsuitable for mainland Chinese adolescents. Indeed, results suggested that three
factors better represented the underlying structure of CES-D. Notably, the resultant best
fitting, three-factor structure, intertwined DA and SC items. In addition, results suggested the
best fitting model was one which deleted two of the CES-D items. Furthermore, the
three-factor structure was shown to be invariant across subgroups of boys and girls. Therefore,
the current study provides further evidence that boys and girls have a uniform structure of
depression symptoms, in other words no significant gender differences were found.
Factor structure and invariance of the CES-D 20
Acknowledgments:
This study is supported by grants from the National Key Technologies R&D Program in the
11th 5-year plan of China (grant no. 2009BAI77B02) and partly by a grant (in part) from
Guangdong Province, philosophy and social sciences "Second Five" planning project (grant
no.GD11CXL03) and partly by the 12th 5-year plan of educational sciences of Hunan
province, (grant no. XJK011BXL002).
Factor structure and invariance of the CES-D 21
Reference
Chan, B., & Parker, G. (2004). Some recommendations to address depression in Chinese
people in Australasia. Australian and New Zealand Journal of Psychiatry, 38, 141–147.
Cheung, C., & Bagley, C. (1998). Validating an American scale in Hong-Kong: The Center
for Epidemiological Studies Depression scale (CES-D). The Journal of Psychology, 132,
169–186. doi:10.1080/00223989809599157
Cheng, C. P., Yen, C. F., Ko, C. H., & Yen, J. Y.(2012). Factor structure of the center for
epidemiologic studies depression scale in Taiwanese adolescents. Comprehensive
Psychiatry, 53, 299-307. doi:10.1016/j.comppsych.2011.04.056
Cheung, F. M. (1995). Facts and myths about somatization among the Chinese. In T.-Y. Lin,
W. S. Tseng, & E. K. Yeh (Eds.), Chinese societies and mental health (pp. 156–180).
Hong Kong: Oxford University Press.
Cheung, G. W., & Rensvold, R. B. (1999). Testing factorial invariance across groups: A
reconceptualization and proposed new method. Journal of Management, 25, 1–27.
doi:10.1177/014920639902500101
Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing
measurement invariance. Structural Equation Modeling, 9, 233–255. doi:10.1207/S15
328007SEM0902_5
Chung, K. F., &Wong, M. C. (2004). Experience of stigma among Chinese mental health
patients in Hong Kong. Psychiatric Bulletin, 28, 451–454. doi:10.1192/pb.28.12.451
Clark, V. A., Aneshensel, C. S., Frerichs, R. R., Morgan, T. M. (1981). Analysis of effects of
sex and age in response to items on the CES-D scale. Psychiatry Research, 124, 171-81.
doi:10.1016/0165-1781(81)90047-0
Crockett, L., Randall, B., Shen, Y., Russell, S., & Driscoll, A. (2005). Measurement
equivalence of the Center for Epidemiological Studies Depression Scale for Latino and
Factor structure and invariance of the CES-D 22
Anglo adolescents: A national study. Journal of Consulting and Clinical Psychology, 73,
47–58. doi:10.1037/0022-006X.73.1.47
DiStefano, C. (2002). The impact of categorization with confirmatory factor analysis.
Structural Equation Modeling, 9, 327-346. doi:10.1207/S15328007SEM0903_2
Edman, J. L., Danko, G. P., Andrade, N., McArdle, J. J., Foster, J., & Glipa, J.(1999). Factor
structure of the CES-D (Center for Epidemiologic Studies Depression Scale) among
Filipino-American adolescents. Social Psychiatry and Psychiatric Epidemiology, 34,
211–215. doi:10.1007/s001270050135
Edwards, M. C., Cheavens, J. S. Heiy, J. E., & Cukrowicz, K. C. (2010). A Reexamination of
the Factor Structure of the Center for Epidemiologic Studies Depression Scale: Is a
One-Factor Model Plausible? Psychological Assessment, 22, 711–715. doi:
10.1037/a0019917
Fan, X., & Sivo, S. A. (2009). Using ΔGoodness-of-Fit Indexes in Assessing Mean Structure
Invariance. Structural Equation Modeling, 16, 54–69. doi: 10.1080/10705510802561311
Flora, D. B., & Curran, P. J. (2004). An empirical evaluation of alternative methods of
estimation for confirmatory factor analysis with ordinal data. Psychological Methods, 9,
466–491. doi:10.1037/1082-989X.9.4.466
Foley, K. L., Reed, P. S., Mutran, E. J., & DeVellis, R. F. (2002).Measurement adequacy of
the CES-D among a sample of older African-Americans. Psychiatry Research, 109,
61–69. doi:10.1016/S0165-1781(01)00360-2
Golding, J., & Aneshensel, C. (1989). Factor structure of the Center for Epidemiologic
Studies Depression Scale among Mexican Americans and non-Hispanic Whites.
Psychological Assessment, 1, 163–168. doi:10.1037//1040-3590.1.3.163
Graham, J.W. (2009).Missing Data Analysis: Making It Work in the Real World. Annual
Review of Psychology, 60, 549–576. doi:10.1146/annurev.psych.58.110405.085530
Factor structure and invariance of the CES-D 23
Greenberger, E., Chen, C., Tally, S.R., & Qi, D. (2000). Family, peer, and individual
correlates of depressive symptomatology among U.S. and Chinese adolescents. Journal
of Consulting and Clinical Psychology, 68, 209–219. doi:10.1037//0022-006X.68.2.209
Guarnaccia, P. J, Angel, R., & Worobey, J. L. (1989). The factor structure of the CES-D in the
Hispanic Health and Nutrition Examination Survey: The influences of ethnicity, gender
and language. Social Science and Medicine, 29, 85-94.
doi:10.1037//0022-006X.68.2.209
Hertzog, C., Van Alstine, J., Usala, P. D., Hultsch, D. F., & Dixon, R. (1990). Measurement
properties of the Center for Epidemiological Studies Depression Scale (CES–D) in older
populations. Psychological Assessment, 2, 64–72.
Iwata, N., Tuner, R.J., Lloyd, D.A. (2002). Race/ethnicity and depressive symptoms in
community-dwelling young adults: a differential item functioning analysis. Psychiatry
Research, 110, 281–289. doi:10.1016/S0165-1781(02)00102-6
Kazarian, S. S.. & Taher, D.(2010). Validation of the Arabic Center for Epidemiological
Studies Depression (CES-D) Scale in a Lebanese Community Sample. European
Journal of Psychological Assessment, 26, 68–73. doi: 10.1027/1015-5759/a000010
Kleinman, A. (1982). Neurasthenia and depression: A study of somatization and culture in
China. Culture, Medicine and Psychiatry, 6, 117–190. doi:10.1007/BF00051427
Kline, R. B. (2010). Principles and practice of structural equation modeling (3rd ed.). New
York, New York: Guilford Press.
Kim, G., DeCoster, J., Huang, C., & Chiriboga, D. A.(2011). Race/Ethnicity and the Factor
Structure of the Center for Epidemiologic Studies Depression Scale: A Meta-Analysis.
Cultural Diversity and Ethnic Minority Psychology, 17, 381–396. doi:
10.1037/a0025434
Kuo, W. H. (1984). Prevalence of depression among Asian-Americans.Journal of Nervous
Factor structure and invariance of the CES-D 24
and Mental Disease, 172, 449–457. doi:10.1097/00005053-198408000-00002
Lam, T. H., Stewart, S. M., Leung, G. M., Lee, P. W. H., & Wong, J. P. S. (2004). Depressive
symptoms among Hong Kong adolescents: Relation to atypical sexual feelings and
behaviors, gender dissatisfaction, pubertal timing, and family and peer relationships.
Archives of Sexual Behavior, 33, 487–496. doi:10.1023/B:ASEB.0000037429.79394.c6
Lee, S. W., Stewart, S. M., Byrne, B. M., Wong, J., Ho, S. Y., Lee, P., & Lam, T. H.(2008).
Factor Structure of the Center for Epidemiological Studies Depression Scale in Hong
Kong Adolescents. Journal of personality assessment, 90, 175-184.
doi:10.1080/00223890701845385
Leykin, Y., Torres, L. D., Aguilera, A., & Muñoz, R. F. (2011). Factor structure of the CES-D
in a sample of Spanish- and English-speaking smokers on the Internet. Psychiatry
Research, 185, 269–274. doi:10.1016/j.psychres.2010.04.056
Liang, J., Van Tran, T., Krause, N., & Markides, K. (1989). Generational differences in the
structures of the CES-D in Mexican Americans. Journal of Gerontology, 44, 110–120.
Li, H. C. W., Chung, O. K. J., & Ho, K. Y.(2010).Center for Epidemiologic Studies
Depression Scale for Children: psychometric testing of the Chinese version. Journal of
Advanced Nursing, 66, 2583-2591.
Lin, Y., Wei, Y., Yi, J., Xiao, J., & Yao, S.(2008). Factorial Structure of the CES-D Scale
Among Chinese High School Students. Chinese Journal of Clinical Psychology, 16,
265-267. (In Chinese).
Lubke, G. H., & Muthen, B. O. (2004). Applying Multigroup Confirmatory Factor Models for
Continuous Outcomes to Likert Scale Data Complicates Meaningful Group
Comparisons. Structural Equation Modeling, 11, 514-534. doi:10.1207/s15328007sem
1104_2
Manson, S. M., Ackerson, L. M., Dick, R. W., Baron, A. E., & Fleming, C. M. (1990).
Factor structure and invariance of the CES-D 25
Depressive symptoms among American Indian adolescents: Psychometric characteristics
of the Center for Epidemiologic Studies Depression Scale (CES–D). Psychological
Assessment, 2, 231–237.
Meade, A. M., Lautenschlager, G. J., & Hecht, J. E. (2005). Establishing measurement
equivalence and invariance in longitudinal data with item response theory. International
Journal of Testing, 5, 279-300. doi:10.1207/s15327574ijt0503_6
Meade, A. W., Johnson, E. C., & Braddy, P. W. (2008). Power and sensitivity of alternative fit
indices in tests of measurement invariance. Journal of Applied Psychology, 93, 568-592.
doi: 10.1037/0021-9010.93.3.568
Meredith, W., & Teresi, J. A. (2006). An essay on measurement and factorial invariance.
Medical Care, 44(11, Suppl 3), S69-S77. doi: 10.1097/01.mlr.0000245438.73837.89
Muthén, L. K., & Muthén, B. O. (1998–2007). Mplus user’s guide (5th ed.). Los Angeles,
California: Muthén & Muthén.
Parker, G., Cheah, Y. C., & Roy, K. (2001). Do the Chinese somatize depression? A
cross-cultural study. Social Psychiatry and Psychiatric Epidemiology, 36, 287–293.
Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the
general population. Applied Psychological Measurement, 1, 385-401.
doi:10.1177/014662167700100306
Radloff, L. S. (1991). The use of the Center for Epidemiologic Studies Depression-Scale in
adolescents and young adults. Journal of Youth and Adolescents, 20, 149-166.
doi:10.1007/BF01537606
Raftery, A. E. (1995). Bayesian model selection in social research. Sociological Methodology,
25, 111–163. doi:10.2307/271063
Roberts, R., E. Andrews, J. A., Lewinsohn, P. M., & Hops, H.(1990).Assessment of
Depression in Adolescents Using the Center for Epidemiologic Studies Depression Scale.
Factor structure and invariance of the CES-D 26
Psychological Assessment, 2, 122-128.
Roberts, R. E., & Vernon, S. W. (1983). The Center for Epidemiologic Studies Depression
Scale: its use in a community sample. American Journal of Psychiatry, 140, 41-46.
Rozario, P. A., & Menon, N.(2010).An examination of the measurement adequacy of the
CES-D among African American women family caregivers. Psychiatry Research, 179,
107–112. doi:10.1016/j.psychres.2010.06.022
Rivera-Medina, C. L., Caraballo, J. N., Rodrıguez-Cordero, E. R., Bernal, G., &
Davila-Marrero, E. (2010). Factor Structure of the CES-D and Measurement Invariance
across Gen for Low-Income Puerto Ricans in a Probability Sample. Journal of
Consulting and Clinical Psychology, 78, 398–408. doi: 10.1037/a0019054.
Ryder, A. G., Bean, G., & Dion, K. L. (2000). Caregiver responses to symptoms of first-onset
psychosis: A comparative study of Chinese- and Euro-Canadian families. Transcultural
Psychiatry, 37, 255–265. doi:10.1177/136346150003700207
Ryder, A. G., Yang, J., Zhu, X., Yao, S., Yi, J., Heine, S. J., & Bagby, R. M.(2008). The
Cultural Shaping of Depression: Somatic Symptoms in China, Psychological Symptoms
in North America? Journal of Abnormal Psychology, 117, 300–313. doi
0.1037/0021-843X.117.2.300
Schroevers, M. J., Sandersman, R., van Sonderen, E., Ranchor, A. V. (2000). The evaluation
of the Center for Epidemiological Studies-Depression (CES-D) scale: depressed and
positive affect in cancer patients and health reference subjects. Quality of Life Research ,
9, 1015–1029.
Shafer, A. B. (2006). Meta-analysis of the factor structures of four depression questionnaires:
Beck, CES-D, Hamilton, and Zung. Journal of Clinical Psychology, 62, 123–146.
doi:10.1002/jclp.20213
Tepper, P., Liu, X., Guo, C., Zhai, J., Liu, T., & Li, C. (2008). Depressive symptoms in
Factor structure and invariance of the CES-D 27
Chinese children and adolescents: Parent, teacher, and self reports. Journal of Affective
Disorders, 111, 291–298. doi:10.1016/j.jad.2008.03.013
Thorson, J., & Powell, F. (1993). The CES-D: Four of five factors? Bulletin of Psychonomic
Society, 31, 577–578.
Yen, S., Robins, C. J., & Lin, N. (2000). A cross-cultural comparison of depressive symptom
manifestation: China and the United States. Journal of Consulting and Clinical
Psychology, 68, 993–999. doi:10.1037//0022-006X.68.6.993
Ying, Y. (1988). Depressive symptomatology among Chinese-Americans as measured by the
CES-D. Journal of Clinical Psychology, 44, 739-746.
doi:10.1002/1097-4679(198809)44:5
Ying, Y., Lee, P. A., Tsai, J. L., Yeh, Y-Y., & Huang, J. S. (2000). The conception of
depression in Chinese American College Students.Cultural Diversity & Ethnic Minority
Psychology, 6, 183–195. doi: 10.1037/1099-9809.6.2.183
Zhang, B., Fokkema, M., Cuijpers, P., Li, J., Smits, N., & Beekman, A. (2011). Measurement
invariance of the center for epidemiological studies depression scale (CES-D) among
chinese and dutch elderly. BMC Medical Research Methodology, 11, 74.
doi:10.1186/1471-2288-11-74
Zhang, J., Wu, Z., Fang, G., Li, J., Han, B., & Chen, Z.(2010). Development of the Chinese
age norms of CES-D in urban area. Chinese Mental Health Journal, 24, 139-143.(In
Chinese).
Factor structure and invariance of the CES-D 28
Table1.
Item Mapping for Tested Models
Item content M1 M2 M3 M4 M5
1. Bothered by things. SC DA DA SC DA
2. Appetite was poor. SC DA DA SC DA
3. Can’t shake off the blues. DA DA DA SC DA
4. Just as good as others. PA DA PA PA PA
5. Trouble concentrating. SC DA DA SC DA
6. Felt depressed. DA DA DA SC DA
7. Everything was an effort. SC DA DA SC DA
8. Hopeful about the future. PA DA PA PA PA
9. Life has been a failure. DA DA DA SC DA
10. Fearful. DA DA DA SC DA
11. Sleep was restless. SC DA DA SC DA
12. Happy. PA DA PA PA PA
13. Talked less than usual. SC DA DA DA DA
14. Felt lonely. DA DA DA DA DA
15. People were unfriendly. IP IP DA DA IP
16. Enjoyed life. PA DA PA PA PA
17. Had crying spells. DA DA DA DA DA
18. Felt sad. DA DA DA DA DA
19. People disliked me. IP IP DA DA IP
20. Could not get going. SC DA DA DA DA
Note. DA = Depressed affect, IP = Interpersonal problem, PA = Positive affect, SC = Somatic
complaints. M1 = Radloff’s original four-factor model; M2 = Two factor model indentified in
Cheung and Bagley’ study (1998), in which DA + PA + SC as a factor and IP; M3 = another
2-factor model in which combined all negative items into a independent factor and remaining
four positive items formed a second factor; Model 4 = positive affect and two new factors
merged from original depressed affect, interpersonal problem, and somatic complaints;
Model 5 = another three-factor model with depressed affect and somatic complaints, positive
affect, and interpersonal problem.
Factor structure and invariance of the CES-D 29
Table 2.
Factor Loadings for EFA in sample 1.
Item content SC DA PA
1. Bothered by things. .698 .170 -.015
2. Appetite was poor. .481 .091 .096
3. Can’t shake off the blues. .664 .249 .179
4. Just as good as others. -.138 .127 .604
5. Trouble concentrating. .666 .104 .033
6. Felt depressed. .673 .365 .176
7. Everything was an effort. .594 .272 .216
8. Hopeful about the future. .111 .064 .713
9. Life has been a failure. .473 .378 .376
10. Fearful. .562 .388 .088
11. Sleep was restless. .454 .223 .035
12. Happy. .271 .136 .694
13. Talked less than usual. .156 .514 .165
14. Felt lonely. .363 .613 .180
15. People were unfriendly. .167 .753 .172
16. Enjoyed life. .223 .139 .732
17. Had crying spells. .260 .493 -.150
18. Felt sad. .559 .503 .059
19. People disliked me. .219 .767 .155
20. Could not get going. .236 .616 .216
Explained Variance 19.89% 16.95% 11.74%
Note. DA = Depressed affect, PA = Positive affect, SC = Somatic complaints.
Depressive Symptom In Chinese Adolescent 30
Table 3
Goodness-of-Fit Indices and Model Comparisons for Tested Models
Model WLSMVχ2 df p TLI CFI BIC RMSEA (90% CI)
M1 2183.48 164 <0.01 .949 .956 99875.173 .070 [.068 -.071]
M2 5350.953 169 <0.01 .873 .887 100941.619 .110 [.108 -.113]
M3 2817.744 169 <0.01 .935 .942 100387.025 .079 [.076 -.081]
M4 2261.602 167 <0.01 .948 .954 99987.883 .070 [ 068 -.073]
M4-modified 1866.193 165 <0.01 .957 .963 99797.885 .064 [.061 -.066]
M5 2179.399 167 <0.01 .950 .956 99862.198 .069 [.066 -.072]
M6 1354.121 132 <0.01 .961 .966 91410.876 .060 [.058 -.063]
Note: CFI = Comparative Fit Index; TLI = Tucker-Lewis index; BIC = Bayesian information criterion; RMSEA = Root Mean Square Error of
Approximation; CI = Confidence interval. M1 = Radloff’s original four-factor model; M2 = Two factor model indentified in Cheung and Bagley’
study (1998), in which DA + PA + SC as a factor and IP; M3 = another 2-factor model in which combined all negative items into a independent
factor and remaining four positive items formed a second factor; M4 = positive affect and two new factors merged from original depressed affect,
interpersonal problem, and somatic complaints; M4-modified = Based on M 4 and allow item 9 and item 18 cross-loading; M5 = another
three-factor model with depressed affect and somatic complaints, positive affect, and interpersonal problem. M6 = Based on M4 but deleted item
9 and item 18.
Depressive Symptom In Chinese Adolescent 31
Table 4
Goodness-of-Fit Indices and Model Comparisons for Measurement Invariance Models
Model WLSMVχ2 df TLI CFI RMSEA (90% CI) Δχ2 Δdf ΔTLI ΔCFI
A-Configural invariance 2740.835** 264 .958 .963 .061 [.059 -, .063]
B-Weak invariance 2849.063** 279 .958 .962 .060 [.058 - .062] B vs. A 169.341** 15 .000 -.001
C-Scalar invariance 3233.528** 312 .957 .958 .061 [.059 - .063] C vs. B 466.026** 33 -.001 -.004
D-Error Variance invariance 2942.618** 330 .961 .964 .056 [.054 - .058] D vs. C 91.854** 18 +.004 +.006
E-Variances-covariances invariance 2520.412** 318 .967 .969 .052 [.050 - .054] E vs. C 14.574* 6 +.006 +.005
F-Latent mean invariance 2512.446** 321 .968 .969 .052 [.050 - .054] F vs. E 55.978** 3 +.001 .000
Note: * p < .05, ** p < .01; CFI = Comparative Fit Index; TLI = Tucker-Lewis index; RMSEA = Root Mean Square Error of a Approximation;
CI = Confidence interval. Δχ2= Change in χ2 relative to the preceding model; Δdf = Change in degree of freedom relative to the preceding model;
∆CFI = Change in comparative fit index relative to the preceding model; ∆TLI = Change in Tucker-Lewis index relative to the preceding model.
Chi-square difference test with WLSMV estimation is different from the conventional chi-square difference test; more information on this can be
found at: http://www.statmodel.com/chidiff.shtml.